AMC-NLI:基于实体识别的农业测控领域自然语言接口  

AMC-NLI:A natural language interface for agricultural measurement and control based on entity recognition

作　　者：袁伟皓 齐海燕 杨梦道 许高建[1] YUAN Weihao;QI Haiyan;YANG Mengdao;XU Gaojian(School of Information and Artificial Intelligence,Anhui Agricultural University,Hefei 230036,China)

机构地区：[1]安徽农业大学信息与人工智能学院,合肥230036

出　　处：《农业工程学报》2024年第19期114-123,共10页Transactions of the Chinese Society of Agricultural Engineering

基　　金：安徽省高校自然科学研究重点项目(KJ2020A0106);安徽省大学生创新创业计划资助项目(S202310364126)。

摘　　要：农业测控系统的用户交互性存在改进空间,随着自然语言语义处理技术的不断进步,提升农业测控领域中复杂的控制和查询操作的用户友好性变得至关重要,这有助于降低用户的操作成本。本文提出了一种面向农业测控领域的自然语言接口(agricultural measurement and control natural language interface,AMC-NLI),旨在改进农业测控平台的用户体验。通过BERT-BiLSTM-ATT-CRF-OPO(bidirectional encoder representations from transformers-bi-directional long shortterm memory-attention-conditional random field)的语义解析模型,识别并提取农业指令中的实体,并进行操作-地点-对象三元组语句(operate-place-object,OPO)的槽填充。使得用户的自然语言输入能够被转化为结构化的三元组语句,实现用户输入的指令转换为相应的参数,并通过物联网网关发送到相应的设备。试验结果表明在AMC-NLI农业测控指令交互方面,该模型表现出色,准确率,精确率、召回率,F值和平均最大响应时间分别达到了91.63%、92.77%、92.48%、91.74%和2.45 s,为农业信息化管控提供了更为便捷的互动方式。User interactivity can be enhanced in agricultural measurement and control systems,especially with the continuous advancements in natural language semantic processing.It is necessary to improve user-friendliness in control and query operations within the agricultural measurement and control field,in order to reduce the user operating costs.Firstly,a precise interface of human-computer interaction can be constructed to tailor for the agricultural domain,in order to efficiently translate the user's natural language input into understandable commands for the computer system.The current agricultural field has relied mainly on graphical user interfaces to meet human-computer interaction.But some limitations still remained over time,e.g.,the high complexity of human-computer interaction and the low efficiency.Therefore,natural language interface(NLI)has been designed to establish the mapping between natural language from the nature of human-computer interaction.Agricultural measurement and control systems have been considered as the efficient strategy.Among them,the primary task of natural language understanding(NLU)is often used to transform the human language into computer-understandable structured expressions,in order to accurately capture the user's intention and semantics.Deep learning has been utilized to name entity recognition tasks in recent years.Relational components of sentences can be extracted to identify the sentence actions,and then incorporate the annotations of semantic roles,in order to understand the utterances for the computers.Entity recognition has distinctly realized the entity features in the specific domains.Commonly-named entities are usually characterized by fuzzy boundaries in the field of agricultural measurement and control systems.Some challenges remain in the quality of data and the accuracy of annotations,due to the relatively scarce data.It is important to directly apply to the agricultural measurement and control system.In this study,the agricultural measurement and control natural l

关 键 词：指令 解析 处理 自然语言 人机交互 命名实体识别 农业测控 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]